November 17, 2006

Responses to troop numbers posts

Some of the feedback I received from readers on the troop numbers posts, below the fold.

An anonymous emailer writes:

I'm not a military analyst, but I was working at the US embassy in Baghdad during the period under discussion, took a keen interest in the military picture, and have a reaction to the discussion.

I think there is a fairly simple explanation for the two statistical observations you are trying to explain - the troop increase/violence decrease in two periods, November 2004 - February 2005 and September - November 2005.

Both these periods saw the relatively rare phenomenon of sustained and significant Coalition offensive operations, in both cases to "set the conditions" for electoral events (first election in January 2005, constitutional referendum in October 2005, election for permanent government in December 2005). Of course the troop plus-ups were specifically for these offensive initiatives, but it's the kinetic operations themselves that were the point.

When on offense, with the exception of unusual cases like Fallujah II in November 2004 against a comparatively prepared defense, the Coalition is likely to suffer fewer casualties and also suppress the level of violence because offensive action disrupts enemy activities, kills and captures enemy personnel, and generally puts pressure on the enemy, making it much harder for the enemy to attack Coalition forces at times or places of his choosing.

To simplify it a bit, it's what you do with the troops, not how many are in-theater.

If one were to examine violence levels in specific areas of Iraq to see correlations with Coalition offensive operations, I have no doubt the link would be clear. There aren't many Coalition offensives, unfortunately, but probably enough to make the fairly obvious linkage clear. Hell, there doesn't even have to be an "offensive" - in some case just taking SOME action will have an effect. I still recall that an immediate effect of the hasty blockade thrown up around Fallujah as a result of the contractor incident in April 2004 was a dramatic decrease in car bombs in Baghdad. I was still in the US at that point, and I seem to recall street vendors in Baghdad making this astute intel observation to western reporters - that's how obvious it was.

There's no non-military solution to the Sunni insurgency in Iraq. But the military and civilian leadership has convinced themselves there's no military solution, so while overall success is still likely in the long run, it'll be a longer run than it had to be.

Another writes:

Although this is harder to quantify, it would also be worth seeing whether there's any change in ROE or procedures during periods such as before elections. We have put more troops in theater at such times, but other changes - such as less stringent rules of engagement, or more attention to certain groups or individuals - might account for a difference in enemy successes and failures at such times.

I also notice that you're graphing number of coalition casualties. Does anything different happen to the numbers if you graph number of attacks instead of total casualties? What if you include Iraqi Army/Police. or Iraqi civilian casualties? There might be a tendency to focus attacks on our troops when more are available - or there might be an opposite tendency, a feeling that troops are too strong and attacking civilians would be preferable.

In a similar vein, Jared writes:

Why are you using coalition troop fatalities as the metric for violence in Iraq? Why not use civilian deaths or total violent incidents? Lots of reasons to do that, not the least of which you already pointed out in that the delta in troop numbers is small. I would also argue that the number of troop deaths is pretty small compared to total number of attacks (either against troops or civilians) which possibly makes it less useful.

Have you considered also adding in total security force size by including the growing numbers of Iraqi forces?

(BruceR: see my latest post for my reasons not to use attacks or Iraqi security force stats.)

Charles M. writes:

You may be aware that there is a literature in economics regarding the impact of the number of US policemen on domestic crime. The most striking aspect of this research is the following correlation: cities with more policemen (per capita) have more crime. I've seen several honors students beach themselves on the rock of trying to explain the result with stories of how policemen cause crime, but there is of course is a more elegant explanation, namely that the number of policemen is not randomly determined. Cities where crime is more endemic are likely to have larger police forces in an effort to contain said crime, while cities with little endemic crime will choose to have smaller police forces.

If US troops are likewise being allocated in response to seasonal or endemic sources of violence, we would expect to see the same correlation, even if additional troops reduce the violence that would have otherwise occurred. Note that this does not undercut the possibility that higher troop levels encourage violence ... it merely says that a positive correlation cannot distinguish between the two explanations.

Is there then any way to answer the underlying question? Given the limited number of observations and time-series nature of the present data, probably not. One might, however, be able to expand the data from a time series (monthly observations of Iraq) to a panel data set (monthly observations of cities in Iraq). This would not address the endogenous troops concern detailed above, but it does introduce a higher probability of significant results. (It also opens up the whack-a-mole critique, but I believe that's a distinct concern.) To fully address the endogenous troops concern, one would want some story of troops being sent to cities for reasons that are distinct from anticipation of violence. A whimsical example would be if the US were periodically trying to threaten Syria/Iran and flooding border towns with more troops to do so. Unfortunately, I know of no such pseudo-experiments on this margin.

(BruceR: Agreed, correlation is not causation. In the case of the Iraqi civilian stats in my latest post, however, it's difficult to see the mechanism whereby increases in violence would cause decreases in Coalition troop strength, so one might reasonably assume any causation would work in the reverse.)

John C. writes:

My point in a similar article I wrote at drumwaster,.com is that the violence follows it's own pattern, losely associated with our elections (both primary and other) and news cycles. Since Ramadan is celebrated according to the lunar calendar, it progresses around the Gregorian calendar, doesn't it?

(BruceR: this is a shortcoming of the method I should have noted: what I did try to note that even the fall violent period is probably partly a religious and partly a climatic cycle.)

Finally, there is a fairly lengthy post here I should also respond to:

All the raw data was drawn from the latest Brookings Institution Iraq index, here, and the site, here. I've also in the latest post, imposed some trendline bars and R-squared values, partly because this is the first time I could see any trends that were obviously significant. As to month-lags, I agree, which is why I did all the rate-of-change scattergraphs with a subsequent-month ("2mo") set of points to see if that kind of change would be evident. Except in the case identified in the most recent post below, I haven't found any correlation yet on that basis.

Posted by BruceR at 05:01 PM

Effects of troop numbers in Iraq, part 3

See previous entries here and here.

An obvious criticism of the method used below to estimate the effect of troop increases in Iraq is the choice of Coalition fatalities or casualties as the measure of violence. Even in the best of circumstances, all it is giving us is a measure of the degree of anti-Western violence, which is not the only component of violence in Iraq by any stretch. Aggregate violence in Iraq could conceivably rise or fall independently of the rate with which Western troops are being attacked, under a number of possible scenarios. If the problem to be solved is achieving a stable Iraq, then violence against Coalition troops may not be a particularly useful performance measure.

Measures of violence against Iraqi security forces is not particularly useful either; in addition to seasonal variation, one would need to control for the numbers of security forces available, which has grown rapidly, at least on paper, over the last few years. Weighting the relative value of different kinds of security forces is also an issue, as is the reliability of (possibly self-serving) estimates of the numbers of Iraqis on duty.

Which leaves measures of violence against civilians as what we really want to achieve here. But as most people are aware, this has been a subject of much controversy, particularly over recent estimates made in the Lancet medical journal. Iraq Body Count keeps a monthly tally of media-reported non-military deaths, but it is generally presumed to be undercounting by a significant measure. Iraqi health ministry estimates and morgue counts have also been criticized as potentially undercounting the scope of the national violence problem. There is also the question of who is being counted... insurgents, police, victims of crime, excess deaths due to declining quality of life, etc.?

On the other hand, it was hard to me personally to accept that one could not find *any* correlation between the presence of Western soldiers in Iraq and violence statistics, so I tried a couple of the available publicly statistical sets out, and finally found one that provided a strong correlation between force presence and violence. Unfortunately, it's a rather limited set for other reasons, but it does produce at least one additional interesting conclusion to add to what I have written on this topic thus far.

The data set referred to here is the UN Iraq Assistance Mission's tally of deaths due to violence, which they have provided from January 2006 on. Used as well by the Brookings Institution, this statistic does not differentiate between violent deaths due to crime or other causes. It is based on Iraqi health ministry and morgue tallies, and excludes all Iraqi military and security personnel. Its estimate of 19,925 deaths between Jan. 1 and Aug. 31 of this year is significantly higher than the Iraq Body Count figures for the same period, but hardly implausible.

Putting this limited data set on a scattergraph with American troop levels for the same time period shows a significant negative correlation between total troop levels and civilian deaths. It also shows a significant correlation between increases in troop levels and decreases in civilian deaths a month later, as you can see below:

Coalition Force levels (x) vs UNAMI Civilian Fatalities per month (y), Jan-Aug 2006

In the scatter-graph above, the total Coalition force level is shown on the horizontal scale, with the points indicating the monthly tallies of civilian fatalities as indicated by the UNAMI tally. There is evidence, given a very limited data set, for a close-to-linear relation in this eight-month period between civilian fatalities and force strength. Troop decreases occurred in parallel with increases in the violent death rate, and vice versa.

Changes in Coalition Force levels (x) vs changes in UNAMI Civilian Fatality tallies in the same and subsequent months (y), Jan-Aug 2006

This scattergraph shows the effect of changing troop levels has had. Generally there is no effect discernible in the same month as an increase or decrease, but there is a significant and potentially linear relationship between changes in force level and civilian casualties the month after.

Now, again, this is a very limited data set. But let's try to interpret what we can. What's been happening to Coalition force levels in 2006? Here's another graph that may be useful, even if it's otherwise as dodgy as hell, statistically:

Seasonally Adjusted Iraq Civilian Fatality Variations from Mean (two methods combined)

This graph does something that probably never should be done by a statistician with any reputation worth preserving. Prior to using the UNAMI count, the Brookings Institution figures for Iraqi civilian casualties relied on a modified version of the Iraq Body Count estimates, derived by removing all security force fatalities, and all morgue reports due to the potential for double counting, then multiplied by a factor of 1.75 to offset undercounting. That number, Brookings reported, corresponded to roughly half the UNAMI figure (because the UNAMI figures include deaths from crime as well, supposedly) for comparable months.

So, because the date range for the UNAMI figures is so short, you could try to extend its conclusions by taking the Brookings "modified IBC" number, multiplying by a further factor of 2 to bring it into parity, subtracting a seasonal variation figure based on the average of fatalities in the same month, and end up with the graph you see here. (Modified IBC numbers are in pink; UNAMI numbers are in red.)

When you consider all the bias introduced by this manipulation, on top of the known concerns about the original IBC tally method, you get a number that wouldn't seem to be worth much . So the fairly obvious and significant linear trend in Iraqi violence over time that this graph seems to indicate may be plausible but the individual monthly totals would not appear to be valuable for any month-by-month analysis: the implicit error bars on those modified IBC figures, and their tenuous relationship with the UNAMI numbers, are simply too big. (The correlation found with the UNAMI numbers does not extend backward, either: when you do a scattergraph comparing the combined civilian death numbers to troop levels across the whole period you're back to an apparently random, non-correlating distribution again.)

That said, I think this graph does demonstrate at least some basis for the common conventional wisdom that the violent death rate of Iraqi civilians was higher in 2006 than in 2005 or before. And look at those troop levels: the largest drop in the entire occupation period came in January, 2006: minus 26,000, a 14% drop in the size of the occupation force in one month, followed by further declines until June. AFter that, force increases in July and August seem to have had a positive effect on the violence level.

This sudden drop in force strength, three times the size of any other decrease, was relatively uncommented upon at the time: I certainly didn't note it as significant until now. One has to wonder, given the demonstrated correlations previously in this post between civilian violence and lowered troop levels since, whether this may have been a contributing factor to the 2006 violence. It is even possible history may conclude that in the early months of 2006, it was the act of American forces withdrawing too much, too fast, that lifted the lid on Iraqi-on-Iraqi violence.

All that said, what could one conclude would be the result of another American surge deployment, up to 20,000 or so (a 12% increase, or 180,000 total), for about six months, in the absence of any other changes in force posture or strategy. Statistically, this would be the most likely prediction:

1) Coalition force fatalities would be expected to increase proportionately with the number of troops in theatre (a 5-15% increase in total wounded and fatalities per month; perhaps 30-50 additional KIAs and 300-500 WIA over a six month period).'
2) There would probably be an initial, apparently very promising decrease in the number of Iraqi civilians being killed with more troops on the ground able to exert a greater degree of local control. This effect would diminish over time, however;
3) If January of 2005 was any indication, any rapid subsequent decrease in force levels, violence would likely return to Iraq with a vengeance.

This is, it should be noted the same old Vietnam rat trap: needing more and more troops to do less and less. It appears that raising troop levels incrementally (20,000 or less) again might delay the crisis in Iraq, but it cannot by itself result in the kind of transformative change required to forestall it completely. (Effects of a troop increase in the 30-50,000 range, if such a thing were possible, could still be of a different order of effectiveness, however.)

Posted by BruceR at 02:23 PM

Salutin on Borat: sounds about right

I think Rick Salutin gets it pretty much correct in his assessment of the Borat phenomenon today, but I'd take it a little further.

Borat's popularity is very much a function of the American (and Canadian) public saying, "to hell with foreigners, and their countries; they're all asses anyway." It's making fun of the other, so you can disengage from any moral imperative to treat them fairly. (Salutin is right that the movie's success has nothing to do with Americans making coming to terms with their own closed-mindedness, as some would pretend: the majority of the movie's fans simply want to laugh at the Other without consequences. Borat gives them that opportunity.)

The success of a movie like Borat is an indicator of a growing American public desire for neo-isolationism. Jingopunditry's current advocation of a more aerial warfighting strategy (the "more rubble, less trouble" meme) is another.

Posted by BruceR at 09:53 AM